[AutoTuner] Add compare and record (#54668)

Azure-Tang · Caozhou1995 · web-flow · commit 6fe7b5e27cff · 2023-06-20T10:37:30.000+08:00
* add auto tuner

* compare and record module

* revert launch main

* add prune rule

* add unit test

* add auto tuner

* revert launch main

* add prune rule

* modify unit test script

* fix bug for dump nodes; fix bug for checking log file

* fix bug

---------

Co-authored-by: caozhou &lt;caozhou@radi.ac.cn&gt;
diff --git a/python/paddle/distributed/auto_tuner/prune.py b/python/paddle/distributed/auto_tuner/prune.py
@@ -83,6 +83,12 @@ def prune_by_mp(tuner_cfg, cur_cfg, history_cfgs=None):
 
 @register_prune
 def prune_by_pp(tuner_cfg, cur_cfg, history_cfgs=None):
+    """
+    Prune by pp (pipeline-parallelism), the rules are:
+    1. PP degree should be evenly divided by number of layers.
+    2. PP degree should be in the candidates of user defined.
+    3. If no candidates, PP degree should be less than or equal to the number of nodes.
+    """
     pp_degree = cur_cfg.get("pp_degree", None)
     num_layers = tuner_cfg["model_cfg"].get("num_layers", None)
     num_nodes = tuner_cfg.get("num_nodes", 1)
@@ -108,6 +114,12 @@ def prune_by_pp(tuner_cfg, cur_cfg, history_cfgs=None):
 
 @register_prune
 def prune_by_mbs(tuner_cfg, cur_cfg, history_cfgs=None):
+    """
+    Prune by mbs (micro batch size), the rules are:
+    1. Micro batch size should be evenly divided by the local batch size.
+    2. Micro batch size should be in the candidates of user defined.
+    3. Prune if a similar configuration with a larger micro batch size resulted in a valid run.
+    """
     micro_batch_size = cur_cfg.get("micro_batch_size", None)
     global_batch_size = tuner_cfg["model_cfg"].get("global_batch_size", None)
     if global_batch_size:
@@ -146,6 +158,13 @@ def prune_by_mbs(tuner_cfg, cur_cfg, history_cfgs=None):
 
 @register_prune
 def prune_by_sharding(tuner_cfg, cur_cfg, history_cfgs):
+    """
+    Prune by sharding parameters, the rules are:
+    1. Sharding stage and sharding degree should be specified.
+    2. Sharding stage and degree should be in the candidates of user defined.
+    3. If PP (pipeline-parallelism) degree is not 1, sharding stage must be 1.
+    4. Prune if a similar configuration with a lower sharding stage resulted in a valid run.
+    """
     sharding_stage = cur_cfg.get("sharding_stage", None)
     sharding_degree = cur_cfg.get("sharding_degree", None)
     pp_degree = cur_cfg.get("pp_degree", None)
@@ -183,11 +202,24 @@ def prune_by_sharding(tuner_cfg, cur_cfg, history_cfgs):
                 and cfg.get("time", -1) > 0
             ):
                 return True
+
+    if sharding_degree == 1:
+        cfgs = same_cfgs_beside("sharding_stage", cur_cfg, history_cfgs)
+        if cfgs:
+            return True
+
     return False
 
 
 @register_prune
 def prune_by_recompute(tuner_cfg, cur_cfg, history_cfgs):
+    """
+    Prune by recompute parameters, the rules are:
+    1. If recompute is not used, return False directly.
+    2. Usage of recompute and recompute granularity should be in the candidates of user defined.
+    3. If recompute is not used, but recompute granularity is set, return True for pruning.
+    4. Prune if a similar configuration without using recompute resulted in a valid run.
+    """
     recompute_granularity = cur_cfg.get("recompute_granularity", None)
     use_recompute = cur_cfg.get("use_recompute", None)
     if not use_recompute:
@@ -221,6 +253,10 @@ def prune_by_recompute(tuner_cfg, cur_cfg, history_cfgs):
             ):
                 return True
 
+    if use_recompute is False:
+        cfgs = same_cfgs_beside("recompute_granularity", cur_cfg, history_cfgs)
+        if cfgs:
+            return True
     return False
 
 
diff --git a/python/paddle/distributed/auto_tuner/recorder.py b/python/paddle/distributed/auto_tuner/recorder.py
@@ -0,0 +1,80 @@
+# Copyright (c) 2023 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     https://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import csv
+import os
+from typing import Tuple
+
+import pandas as pd
+
+
+class History_recorder:
+    # NOTE increase extenable ablitity
+    def __init__(self) -> None:
+        self.history = []
+        self.store_path = None
+
+    def add_cfg(self, **kwargs):
+        cur_configs = {}
+        for key, val in kwargs.items():
+            cur_configs[key] = val
+        self.history.append(cur_configs)
+
+    def sort_metric(self, direction, metric_name) -> None:
+        if direction == 'Maximize':
+            self.history.sort(
+                key=lambda x: x[metric_name]
+                if x[metric_name] is not None
+                else float('-inf'),
+                reverse=True,
+            )
+        else:
+            self.history.sort(
+                key=lambda x: x[metric_name]
+                if x[metric_name] is not None
+                else float('inf'),
+                reverse=False,
+            )
+        return
+
+    def get_best(self, metric, direction) -> Tuple[dict, bool]:
+        self.sort_metric(direction=direction, metric_name=metric)
+        if len(self.history) == 0:
+            return (self.history[0], True)
+        return (self.history[0], False)
+
+    def store_history(self, path="./history.csv"):
+        """Store history to csv file."""
+        self.store_path = path
+        # convert to pd dataframe
+        df = pd.DataFrame(self.history)
+        # move 'job_id' to the first column
+        cols = df.columns.tolist()
+        cols.insert(0, cols.pop(cols.index('job_id')))
+        df = df.reindex(columns=cols)
+        # write to csv
+        df.to_csv(self.store_path, index=False)
+
+    def load_history(self, path="./history.csv") -> Tuple[list, bool]:
+        """Load history from csv file."""
+        err = False
+        if self.store_path is None:
+            self.store_path = path
+        if not os.path.exists(self.store_path):
+            err = True
+        else:
+            with open(self.store_path, "r") as f:
+                reader = csv.reader(f)
+                self.history = list(reader)
+        return (self.history, err)
diff --git a/python/paddle/distributed/auto_tuner/utils.py b/python/paddle/distributed/auto_tuner/utils.py
@@ -14,6 +14,9 @@
 
 import copy
 import itertools
+import os
+import re
+from typing import Tuple
 
 
 def divisor(num, reverse=False):
@@ -231,3 +234,38 @@ def gen_new_args(raw_args, cfg, tuner_cfg):
         res_args.extend(cmd["local_batch_size"])
 
     return res_args
+
+
+def read_log(
+    path, file="workerlog.0", target_metric='step/s'
+) -> Tuple[float, bool]:
+    """For extracting metric from log file."""
+    target_file = path + "/" + file
+    if not os.path.exists(target_file):
+        return (0.0, True)
+    with open(target_file, "r") as f:
+        # read file
+        re_metric_pattern = r'speed: (\d+(\.\d*)?) *' + target_metric
+
+        metric_list = []
+        lines = f.readlines()
+        for line in lines:
+            metric = re.findall(re_metric_pattern, line)
+            if metric:
+                metric_list.append(float(metric[0][0]))
+        if not metric_list:
+            metric_ave = 0.0
+            flag = True
+        elif len(metric_list) < 10:
+            metric_ave = metric_list[-1]
+            flag = False
+        elif len(metric_list) < 20:
+            metric_ave = sum(metric_list[9:]) / (len(metric_list[9:]))
+            flag = False
+        else:
+            metric_ave = sum(metric_list[-10:]) / 10
+            flag = False
+        # round to 5 decimal places
+        metric_ave = round(metric_ave, 5)
+    res = metric_ave, flag
+    return res
diff --git a/python/paddle/distributed/launch/main.py b/python/paddle/distributed/launch/main.py
@@ -299,8 +299,9 @@ def launch():
         import sys
         import time
 
+        from ..auto_tuner.recorder import History_recorder
         from ..auto_tuner.tuner import AutoTuner
-        from ..auto_tuner.utils import gen_new_args
+        from ..auto_tuner.utils import gen_new_args, read_log
         from . import controllers
 
         # read user defined tuner config json
@@ -323,7 +324,11 @@ def launch():
             gpus_per_node = 8
         else:
             gpus_per_node = len(ctx.args.devices.split(","))
-        tuner_cfg["nodes"] = int(ctx.args.nnodes)
+        nnodes = ctx.args.nnodes
+        if isinstance(nnodes, str):
+            tuner_cfg["nodes"] = int(nnodes.split(":")[0])
+        else:
+            tuner_cfg["nodes"] = int(nnodes)
         tuner_cfg["num_gpus"] = gpus_per_node * tuner_cfg["nodes"]
 
         # build AutoTuner to get new config
@@ -333,12 +338,16 @@ def launch():
         # get max time per task run
         max_time_per_task = tuner_cfg.get("max_time_per_task", 1800)
 
+        # build history recorder
+        recorder = History_recorder()
+
         job_id = 0
         while cur_cfg:
+            ctx.status._current_status = None
             # auto tuner supports dp, mp, pp, micro batch size, sharding, recompute by default and every task has own log dir
             log_dir = "DP{}_MP{}_PP{}_Sharding_degree_{}_stage_{}_MBS_{}_Recompute_{}_granularity_{}".format(
                 cur_cfg["dp_degree"],
-                cur_cfg["pp_degree"],
+                cur_cfg["mp_degree"],
                 cur_cfg["pp_degree"],
                 cur_cfg["sharding_degree"],
                 cur_cfg["sharding_stage"],
@@ -371,19 +380,47 @@ def launch():
             signal.alarm(max_time_per_task)
             c.run()
 
+            # Process generated result
+            metric, err = read_log(
+                path=ctx.args.log_dir,
+                file="workerlog.0",
+                target_metric=tuner_cfg["metric_cfg"]["name"],
+            )
+            if err:
+                ctx.logger.warning(f"Read log failed for parameters: {log_dir}")
+                cur_cfg['time'] = None  # for pruner use.
+                cur_cfg[tuner_cfg['metric_cfg']['name']] = None
+            else:
+                cur_cfg['time'] = metric  # for pruner use.
+                cur_cfg[tuner_cfg['metric_cfg']['name']] = metric
+                # record history
+            cur_cfg['job_id'] = job_id
+            recorder.add_cfg(**cur_cfg)
+            cur_best_cfgs, err = recorder.get_best(
+                metric=tuner_cfg['metric_cfg']['name'],
+                direction=tuner_cfg['metric_cfg']['OptimizationDirection'],
+            )
+            if not err:
+                ctx.logger.info(f"Current best config: {cur_best_cfgs}")
+                recorder.store_history(
+                    ctx.args.auto_tuner_json.split(".")[0] + "_history.csv"
+                )
+            else:
+                ctx.logger.info(
+                    "Get best config failed. Currently there are no appropriate configs."
+                )
+
             new_cfg = auto_tuner.search_once()
             if new_cfg:
                 c.finalize(exit=False)
             else:
                 c.finalize(exit=True)
 
-            # NOTE: The statistics and comparison function of task results will be implemented in the future.
-
             # per task launch interval
             time.sleep(5)
 
             cur_cfg = copy.deepcopy(new_cfg)
-
+        recorder.store_history()
     else:
         from . import controllers
 
diff --git a/test/auto_parallel/CMakeLists.txt b/test/auto_parallel/CMakeLists.txt
@@ -68,6 +68,9 @@ if(WITH_DISTRIBUTE AND WITH_GPU)
   py_test_modules(test_auto_tuner MODULES test_auto_tuner)
   set_tests_properties(test_auto_tuner PROPERTIES LABELS "RUN_TYPE=EXCLUSIVE"
                                                   TIMEOUT 100)
+  py_test_modules(test_auto_tuner_compare MODULES test_auto_tuner_compare)
+  set_tests_properties(test_auto_tuner_compare
+                       PROPERTIES LABELS "RUN_TYPE=EXCLUSIVE" TIMEOUT 100)
   # End of unittests WITH multi cards and timeout
 
   # NOTE(zyl): unittests WITH multi cards and WITHOUT timeout
diff --git a/test/auto_parallel/test_auto_tuner.py b/test/auto_parallel/test_auto_tuner.py
@@ -61,6 +61,10 @@ def test_auto_tuner(self):
                 "use_recompute": ["-o", "Model.use_recompute"],
                 "recompute_granularity": ["-o", "Model.recompute_granularity"],
             },
+            "metric_cfg": {
+                "name": "step/s",
+                "OptimizationDirection": "Maximize",
+            },
         }
 
         tmp_dir = tempfile.TemporaryDirectory()
diff --git a/test/auto_parallel/test_auto_tuner_compare.py b/test/auto_parallel/test_auto_tuner_compare.py
